Seasonal forecasting of Ethiopian spring rains

Abstract

AbstractReliable prediction of rainfall in advance of the rainy season would have enormous social and economic benefits to countries such as Ethiopia that depend on rain‐fed agriculture. Moreover, the development of statistical seasonal forecasting models provides insights into important scientific problems such as the predictability of rainfall, the large‐scale controls on rainfall, and the teleconnections with tropical and extra‐tropical processes. This article describes the development of statistical seasonal forecasts of the Ethiopian spring rains. Because of the spatial variation in both the interannual variability and the annual cycle of rainfall, Ethiopia was divided into five homogeneous rainfall zones, and separate forecasts were developed for each zone. Two techniques (multiple linear regression (MLR) and linear discriminant analysis (LDA)) were applied to four sets of predictors (selection by either stepwise regression or discriminant analysis either including or excluding the contemporaneous season). All the forecasts had more skill than either a random or climatological forecast. The method of selecting predictors and developing the models was found to have surprisingly little impact on forecast skill. Interestingly, including contemporaneous SST did not always lead to a significant increase in skill. Comparison between the skill of the forecasts in different zones shows that the models had most skill in the southern and eastern parts of Ethiopia. For all forecasts, the extreme years (very rainy and very dry) were more reliably forecast than the average years. Copyright © 2008 Royal Meteorological Society